AI Literacy Programme

The EU AI Act requires organisations that provide and use AI systems to ensure their staff have a sufficient level of AI literacy. The obligation is independent of risk class: it applies even where your systems are minimal risk.

This page describes how we meet it.

Who takes what

The core module is mandatory for everyone, without exception.

Content

Core literacy — all staff

• What AI does and does not do; hallucination, and the rule that every factual claim is verified at its source
• Automation bias — the tendency to suppress one's own judgement when a machine speaks, and the countermeasure
• Data privacy — customer, card and identity data must never be entered into external AI tools; a concrete list of prohibitions
• Where AI runs on the platform and what decision authority each component holds
• How to report an AI-related problem

Engineering — software and data

• The model lifecycle and its gates: shadow, gradual rollout, adoption; the threshold monitor alerts but does not roll back
• Explainability as a requirement rather than an option: a decision path that produces no rationale is not added
• Raising schema and model version together; rollback preserving learned state
• The boundary on data reaching external model services, and the two questions to answer when adding a model call
• Prompt injection: external text reaching a model is data, not instruction; the side-effecting node palette stays closed
• No personal data in test fixtures

Decisions and oversight — compliance, risk, operations, analysts

• A score is a weight, not a probability; it is read together with its reason codes
• When to challenge a suggestion — an indeterminate signal, a rationale that does not carry the decision, a systematic skew
• The final decision rests with the analyst, and an AI suggestion cannot be recorded as the decision's rationale
• Controller versus processor in automated decisions, and whose obligation the objection process is
• The tipping-off prohibition and the access list on case content

Regulatory responsibility — management

• The risk classification of the platform's components and the reasoning behind it — see EU AI Act
• The obligations that bind us
• Decisions that would break the classification, and that these require management approval

Assessment and record

Each module ends with a short assessment; the pass threshold is 80%. Staff who do not pass repeat the training.

Attendance is recorded per person and date. The record is the evidence requested in an audit and is shared under NDA as part of a vendor assessment — request it at [email protected].

A session for your team

If your compliance, risk or operations team will be using the platform, we can deliver the decisions and oversight module above as a session tailored to your own configuration: your published rules, your reason codes and your case flow.

Request one at [email protected].

What governs the programme itself

Related pages

• Responsible AI Policy — principles, human oversight, explainability, monitoring
• EU AI Act — risk classification and its reasoning
• Model Review Log — the record of scheduled reviews

Questions

For the programme content, a request for the attendance record, or a vendor assessment form: [email protected]